# This is not an AI Art Podcast (Ep. 10)
## Intro
Welcome to episode ten! This is your host, Doug Smith. This is Not An AI art podcast is a podcast about, well, AI ART – technology, community, and techniques. With a focus on stable diffusion, but all art tools are up for grabs, from the pencil on up, and including pay-to-play tools, like Midjourney. Less philosophy – more tire kicking. But if the philosophy gets in the way, we'll cover it.
But plenty of art theory!
Little different this week!
Today we've got:
* Model madness model reviews: Two models
* Let's get weird: Oddities from the world of prompt engineering, and prompt engineering examples
* Technique of the week: Bashing with control net
* My project update: so you can learn from my process
Available on:
* [Spotify](https://open.spotify.com/show/4RxBUvcx71dnOr1e1oYmvV)
* [iHeartRadio](https://www.iheart.com/podcast/269-this-is-not-an-ai-art-podc-112887791/)
* [Google Podcasts](https://podcasts.google.com/feed/aHR0cHM6Ly9hbmNob3IuZm0vcy9kZWY2YmQwOC9wb2RjYXN0L3Jzcw)
Show notes are always included and include all the visuals, prompts and technique examples, the format is intended to be so that you don't have to be looking at your screen -- but the show notes have all the imagery and prompts and details on the processes we look at.
## News
## Model Madness
### Reliberate
[On Civitai](https://civitai.com/models/79754/reliberate)
A "realistic" version of Deliberate, apparently! Which is a GREAT model.
This model is definitely NSFW (like which ones aren't?) it likes to give me a lot of more bare bodies than I prompted for `:sweat-smile:`
But yeah! I think it's really good. I'm getting good results out of it, and it's definitely similar to deliberate -- it can really get a lot of good realistic stuff, but it does shine with some more artistic flair.
Flappers came out OK, but I couldn't get color photography? Huh. So that's boring so I did some more samples. I felt like I was getting some amount of rigity here. But, it could be me.
```
a portrait photograph of a 1920s flapper, speakeasy, 1920s fashion, dim lighting, alluring, looking at viewer, detailed skin, highly detailed, diffused soft lighting, hyperrealism, cinematic lighting, (color photography:1.2)
Negative prompt: (bad_prompt_v2:0.8),Asian-Less-Neg,bad-hands-5, UnrealisticDream, (skinny:1.2), black and white, monochrome
Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 6, Seed: 1082815664, Face restoration: CodeFormer, Size: 576x768, Model hash: 980cb713af, Model: reliberate_v10, VAE: vae-ft-mse-840000-ema-pruned, Denoising strength: 0.5, Hires upscale: 1.5, Hires steps: 10, Hires upscaler: Latent
```
I interrogated this [image on pinterest](https://www.pinterest.com/pin/friends--345229127658076526/) that I found with google images.
```
a raver woman in a yellow shirt and jeans posing for a picture, warehouse rave, dance trance edm festival, colorful pigtail, lunapunk, iowa, her belly button is exposed, trash polka, dayglo pink blue, pale sober colors 9 0 %, bandana, 9 0 - s, minneapolis, dark rainbow colored fur, outfit photograph, one blonde, AS-MidAged
Negative prompt: (bad_prompt_v2:0.8),Asian-Less-Neg,bad-hands-5, UnrealisticDream, (skinny:1.2)
Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 6, Seed: 3969000457, Face restoration: CodeFormer, Size: 576x768, Model hash: 980cb713af, Model: reliberate_v10, VAE: vae-ft-mse-840000-ema-pruned, Denoising strength: 0.5, Hires upscale: 1.5, Hires steps: 10, Hires upscaler: Latent
```
And then image to image
### A-zovya Photoreal
[On Civitai](https://civitai.com/models/57319/a-zovya-photoreal)
Having pretty good luck with it! Prompt against renders, I think is kind of important.
Also having extremely good luck with hands?
I really like the generations I'm getting, this might be the pick of the week.
```
photograph of a 1920s flapper (masterpiece:1.2) (photorealistic:1.2) (bokeh:1.2) (best quality) (detailed skin:1.3) (intricate details) (8k) (HDR) (cinematic lighting) (sharp focus)
Negative prompt: (bad_prompt_v2:0.8),Asian-Less-Neg,bad-hands-5, UnrealisticDream, (skinny:1.2), (render, drawing, anime)
Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 6, Seed: 2606510074, Face restoration: CodeFormer, Size: 576x768, Model hash: f55ed5fbe2, Model: aZovyaPhotoreal_v1VAE, VAE: vae-ft-mse-840000-ema-pruned, Denoising strength: 0.5, Hires upscale: 1.5, Hires steps: 10, Hires upscaler: Latent
```
```
a raver woman in a yellow shirt and jeans posing for a picture, warehouse rave, dance trance edm festival, colorful pigtail, lunapunk, iowa, her belly button is exposed, trash polka, dayglo pink blue, pale sober colors 9 0 %, bandana, 9 0 - s, minneapolis, dark rainbow colored fur, outfit photograph, one blonde, AS-MidAged
Negative prompt: (bad_prompt_v2:0.8),Asian-Less-Neg,bad-hands-5, UnrealisticDream, (skinny:1.2)
Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 6, Seed: 775891595, Face restoration: CodeFormer, Size: 576x768, Model hash: f55ed5fbe2, Model: aZovyaPhotoreal_v1VAE, VAE: vae-ft-mse-840000-ema-pruned, Denoising strength: 0.5, Hires upscale: 1.5, Hires steps: 10, Hires upscaler: Latent
```
And from img2image as before...
## Let's get weird!
### scp anamalous object
```
scp anomalous object
Negative prompt: (bad_prompt_v2:0.8),Asian-Less-Neg,bad-hands-5, UnrealisticDream, (skinny:1.2), (render, drawing, anime)
Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 5, Seed: 1250804271, Face restoration: CodeFormer, Size: 576x768, Model hash: 0b914c246e, Model: analogMadness_v40, VAE: vae-ft-mse-840000-ema-pruned, Denoising strength: 0.5, Hires upscale: 1.5, Hires steps: 10, Hires upscaler: Latent
```
### ice! cream! truck!
```
ice! cream! truck!
Negative prompt: (bad_prompt_v2:0.8),Asian-Less-Neg,bad-hands-5, UnrealisticDream, (skinny:1.2), (render, drawing, anime)
Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 5, Seed: 100, Face restoration: CodeFormer, Size: 576x768, Model hash: 0b914c246e, Model: analogMadness_v40, VAE: vae-ft-mse-840000-ema-pruned
```
### MBTI? Does it do anything?
It's sort of "corporate astrology" kind of thing. I'm not sure how effective it is, but you might want to try it here and there.
```
portrait photograph of woman from vermont, (istj:1.2), RAW photo, analog style, fujifilm XT3
Negative prompt: (bad_prompt_v2:0.8),Asian-Less-Neg,bad-hands-5, UnrealisticDream, (skinny:1.2), (render, drawing, anime)
Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 5, Seed: 100, Face restoration: CodeFormer, Size: 576x768, Model hash: 0b914c246e, Model: analogMadness_v40, VAE: vae-ft-mse-840000-ema-pruned
```
### dance trance edm festival
Got this from the raver photo earlier.
```
dance trance edm festival, RAW photo, analog style, 4k, fujifilm xt3
Negative prompt: (bad_prompt_v2:0.8),Asian-Less-Neg,bad-hands-5, UnrealisticDream, (skinny:1.2), (render, drawing, anime)
Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 5, Seed: 2122981753, Face restoration: CodeFormer, Size: 576x768, Model hash: 980cb713af, Model: reliberate_v10, VAE: vae-ft-mse-840000-ema-pruned, Denoising strength: 0.5, Hires upscale: 2, Hires steps: 10, Hires upscaler: Latent
```
### nineties nostalgia
```
nineties nostalgia
Negative prompt: (bad_prompt_v2:0.8),Asian-Less-Neg,bad-hands-5, UnrealisticDream, (skinny:1.2), (render, drawing, anime)
Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 5, Seed: 3580770512, Face restoration: CodeFormer, Size: 576x768, Model hash: 980cb713af, Model: reliberate_v10, VAE: vae-ft-mse-840000-ema-pruned
```
## Technique of the week: Map Bashing
So I've been doing this now and again, and I just learned there's a name for it, like photobashing (adding photographic elements to your img2img input), this is "map bashing" -- adding portions of one a control net "map" to another map.
You can do this with any map, but I'm finding that I get great results out of the line art system, it's really easy to work with the maps, too.
I used this to create a (somewhat) consistent scene -- at least good enough to represent this landscape scene in a believable enough
I started with this map, based on a postcard:
Then, I generated images, and went to control net line art, and generated maps from them, and then I combined them in my image editor.
I kind of made the prompt template before I had them, so once I had them I knew my formula and generated and upscale a batch about ~25-50 each.
Then I'd pick one and inpaint it, and I wind up with:
## Project update
So on my project, had a really great week between a radio interview and a magazine feature!
### A story on GPS
However, something interesting has happened that I got basically called out for "GPS" in the comments, "the skin is too perfect, just sayin'". We all know why -- it's the dataset and face repair. That's fine, and it's a reality of social media: people can comment on your stuff. I think in part because people are picking up on the fact that this is specifically AI generations, that they feel liberated to make any comment on it. Which is fine, but! It's a challenge compared to say, posting a pencil drawing (generally) no one's going to say "Your foreshortening is fucked up." it's too personal.
So let's call those out first, before someone else can. All the more reason to crit the work.
Two points:
* This detracts from the "suspension of disbelief"
* Compared to say having someone say "Do you have more information on $subject", because it's personally important to me -- which is more the experience I have typically had, along those lines.
* Isn't as good for the narrative, see?
* This is the reason we have to have our shit on lock
* Generally, people are going to wise up on how to recognize this stuff, quickly. You already have, everyone is getting there too
* You gotta get rid of the AI-based look, OR, you have to push that and make it part of the piece.
Sign in with Wallet
Connect another wallet